Re: [9fans] double->string conversion

Re: [9fans] double->string conversion

[rog]

> doesn't print("%g") do it?

nope. for example, 

	double xmin, xmax;
	[...]
	print("%g %g; difference: %g\n", xmin, xmax, xmax - xmin);
	print("hex: %ullx %ullx\n", *(vlong *)&xmin, *(vlong *)&xmax);

prints:

	-.9909985 -.9909985 difference: 4.551914e-15
	hex: bfefb6429008c02c bfefb6429008c003

> The %f, %g, strtod functions in both Inferno and 3rd edition Plan 9
> are intended to be accurate.   (The version in 2nd edition is known
> to be less exact.)

what parameters does the %g conversion need to ensure
that a later strtod will produce the same value?

  rog.

Re: [9fans] double->string conversion

[Eric Grosse]

> what parameters does the %g conversion need to ensure
> that a later strtod will produce the same value?

%.17g is enough.

http://plan9.bell-labs.com/netlib/fp/g_fmt.c.gz gives a
shorter string, in general, that also promises to exactly
reproduce the starting binary value.

Eric

Re: [9fans] double->string conversion

[Eric Grosse]

The %f, %g, strtod functions in both Inferno and 3rd edition Plan 9
are intended to be accurate.   (The version in 2nd edition is known
to be less exact.)

The one in Inferno comes from David Gay and is the more heavily tested;
see http://plan9.bell-labs.com/netlib/fp/.

The one in Plan 9 (/sys/src/libc/port/fltconv.c and strtod.c) is by
Ken Thompson and should give the same conversion.  It is shorter but
probably not as fast.   If you find accuracy problems with either
one, please let me know.

Eric

[9fans] double->string conversion

[rog]

is there a plan 9 interface to an accurate float to string
conversion routine, like inferno's sys->sprint("%g")?

Re: [9fans] double->string conversion

[Russ Cox]

> is there a plan 9 interface to an accurate float to string
> conversion routine, like inferno's sys->sprint("%g")?

doesn't print("%g") do it?